I did not run it (obviously), but it looks perfectly functional.

The only concern that I will mention is one that I have expressed before; I am worried about the inherent non-repeatability/non-reproducibility that comes from reusing the same filenames. (Feel free to consider this asked and answered.)

I'm going to echo @jamesmcclain's comments here. Naming is a problem. I don't recommend SHAs, though, as a replacement. Would rather have something like timestamp so that it's easy to figure out what the last build was.

Also, I don't think that the geopyspark-dependency-jar bucket is my favorite name. I've created a geopyspark-resources bucket that contains an rpms folder. There should probably be a jars folder under that. Collect all the things under one umbrella, but compartmentalized.

Finally, my last concern is that the egg you're building seems to be based on python 3.5. Is that the officially supported python version? Bear in mind that EMR comes pre-installed with 3.4.3. This is what appears to be the case, btw. I might be wrong on this point.